Miscellaneous Operating Systems/Hardware

http://blogs.technet.com/windowsserver/ ... anium.aspx

I do not like this at all, but since Itanium is increasingly losing ground to POWER and that damn x64 architecture, and since most IA64 boxes sold are HP-UX or VMS anyway, I understand where they're coming from.

I tried to load it on my RX2600 but it didn't have like any of the drivers for it lol. Also it's very limited in what it can do compared to the 32 and 64 bit x86 versions.

Sorry, I'm not following you here, you're upset that Windows won't run on your hardware anymore?!

I think the main complaint is the majority global CPU architecture is based on the 4004.

My complaint isn't that I personally love Windows and care that it doesn't run on my own Itanium hardware, but that this is a win for x86/x64. Maybe I'm stupid and obsolete, but I refuse to believe that that architecture is the future or anything close to it...

Unfortunately it seems to be, at least for the short term, but not for technical merit.

98% of the time the underlying architecture is of no consequence to the end user. and much of the time it's of little consequence to the programmer thanks to high-level languages. Money, at least immediate cash outlay, is very evident to the purchaser, and it's hard to argue with AMD64 on a price/performance standpoint.

Perhaps with mobile and ecological awareness ( and economics of electricity bills ) power usage will become a more important consideration.

With all those high-level programming ‘object-oriented miracles’ one would need all the AMD64 CPUs that money can buy... What a wonderful market the whole ICT industry is, isn't it?

I think I'm going to give up buying anything new if the last RISC, or otherwise solid *N*X or VMS-capable and proven, architecture should at some point die off...

I don't understand... electricity is getting cheaper and cheaper to generate and batteries have ever-increasing capacity. re the environment IMO it is better to use something already existing than throw it away and buy something new.

Good point sybrfreq.

The following is on a minimal amount of sleep, and may not be entirely coherent.

Honestly, I don't think you need to worry. SPARC and IA64 are both seemingly at death's door, but POWER is going strong, and plenty of new chipmakers (Tilera, for example) are coming into the market. I would not be surprised to see a "RISC Renaissance" in the high end as personal computing becomes increasingly based on thin-client type technologies (like "cloud computing.") All the architectural band-aids in the world can't make x86's disadvantages go away. The really cost-sensitive segments (desktops and laptops, netbooks, etc) are already about as fast as they need to be, and will probably stick with x86 until something better (ARM?) comes along that can run the same software at a reasonable speed. On the other hand, RISC is still the fastest thing around on severs and large-scale workstations, where performance still matters, and I think that will increasingly shift in RISC's favor. x86 just isn't that fast, doesn't comfortably go above about six cores without MCM's, and in general it's I/O and memory bandwidth don't come close to RISC solutions, especially POWER. Nehalem goes a long way toward correcting this, but getting relatively close in performance to the Power6, which is last-gen, really isn't good enough.

I predict RISC workstations will, in fact, come back. I think personal computing processors over the next few years are going to be more architecturally to embedded processors than to workstation/server processors, and that personal computing will always be where Intel and (probably) AMD do the most R&D. This leaves a hole in workstation and server processors that the fast RISC chips have an excellent chance to fill. I don't think that SiCortex was the last attempt at making RISC workstations; I think it was one of the first of the new generation.

Sorry for all the run-on sentences, bizarre transitions, and non-fleshed-out ideas. I hope I get the point across.

Not from where I'm standing. My electricity bills only go north, similar for server farms. We will be seeing some major constraints powerwise in the not so distant future.

Yes it's better to use what you have than throw away and buy new, but mix that in with a consumer society that requires people to throw away and buy new and you have a problem.

What do you mean by "large scale workstation"? IBM seems to have discontinued their framebuffer equipped POWER boxes (though I have trouble figuring out exactly what they sell rapidly, so I could have missed something).

RISC/VLIW (though they really aren't "RISC" anymore - non-x86 or load/store are better terms) are great technologies, but they are likely to only hold onto the high end of "serious computing" and, possibly, the low-end and client (a la Godson/SunRay/etc.). AMD64 is good enough for what most of the midrange server/commercial machines do (as long as you can put up with the lack of serious RAS facilities) and they're CHEAP! Power-wise they're OK, too - RISC uses lots of power in the gigantic caches usually attached. Now that the high-perf IO and interconnects are filtering down (PCIe/SATA with smarts/HyperTransport and its ilk) AMD64 is just good enough to where it's hard to justify the costs of a POWER system or Itanium, especially when you factor in the costs of a serious OS (OpenVMS, HP-UX or AIX) on the hardware. Linux/xBSD are good, but it's hard to find non-x86 software excluding FOSS.

I'd love to see POWER take over, coupled with a rebirth of Alpha, but I'm not sure that's going to happen anytime soon except for the big boxes (and even then Alpha is going to stay dead).

Sorry for being vague. I meant something like an IntelliStation POWER or medium-size Prism as opposed to the cheap x86 systems or even things like O2's. The fairly large number of SiCortex PDS-72's sold indicate that there's still a market for things like this, even when it's not legacy technology. (try seeing x86 pack 72 cores of that speed into a machine that only consumes 300W!)

Well if it makes you feel better the latest Intel stuff has quite a bit of Alpha tech in it apparently...would be nicer if it was in an actual Alpha though haha.

People were probably still buying them and Intel was pushing HP to drop for everyone to have to switch to Itanium2. PA-RISC definitely have some nice performance numbers for their clock rate.

Yeah, in the late 90s, AMD couldn't just copy the wonderous and soon to be ubiquitous IA64 architecture without licensing it from arch rival Intel, so the only other option was to come up with their own 64-bit x86 compatible solution. Its a little ironic that AMD64 won the 64-bit market, and Intel ended up licensing it for their x86 range (Intel64 mode).

I was teaching OS internals for a while, and had to look at both IA64 and AMD64, as I was initially unsure which would take over from 32-bit x86. IA64 was overengineered and needlessly complex, which hinders adoption by OS and compiler writers. As ugly as 32-bit x86 was, a 64-bit extension of it was more elegant IMO (and far easier to quickly learn and understand). Better the devil you know.


Similar arguments could be made about Alpha. I guess they just had to consolidate resources and decide on one future CPU strategy. At the time, CPU architects such as John Mashey (MIPS) felt that RISC just wouldn't be able to keep up with EPIC. And also, the strategy of teaming up with a chip giant such as Intel for your future CPU needs is probably hard to argue against when you are at your next board meeting.

They wanted someone else to take over a big chunk of the costs associated with processor design/manufacture. It didn't help that HP was generally fabbing their own processors at the time that the decision to go Itanium was made, and fabs were getting more and more expensive. Performance didn't enter into it - Alpha, PA-RISC and MIPS were all at or near the top around the time that the decision to drop them and go with Itanium was made.

Indeed, if SPECmarks were the main category than SPARC (optimized for commercial workloads rather than technical) would have been the likely candidate. It was costs for design and manufacture that did in the custom processors, especially when you couple in CEOs that don't look more than one year down the road (though it's doubtful that the custom designs could be justified even long-term).

Once gate-array processes became too slow to compete things started going up in price exponentially, and you couldn't inexpensively get a design made. Once clock speeds started skyrocketing you had to shrink so often that it was even worse.

All in all I think it's somewhat unfair to compare the IA64 architecture vs the x64 architecture - they're aimed at different markets. Customers who use IA64 inside Integrity machines usually have to rely on multitudes of error-correction schemes to be absolutely dead-sure that their data stays intact and doesn't fsck things up between transactions. Banking, stockmarkets, process industry, etc all rely on their equipment running 100% and can't tolerate failures. This involves very heavy testing in the design phase of these and thorough planning is a must.
Performance for these types of customers doesn't have to be the absolute first priority, neither does TDP values or other stuff - they'll happily pay if they can be sure to keep error-free data. (Is there such things?)

Alot of the blame for the latest Itanium2 chips delay were caused of errors discovered in their 6 month long testing phases they use. (They don't involve more than what? 1 month? on their IA32 stuff) That coupled with requested design features from companies with the big wallets.

What still amazes me is that SGI choose IA64 for their HPC stuff. Ofcourse the IA64 was ment to be the future replacement processor for heavy HPC tasks, but it wasn't a real performer until the later Itanium2 chips came to market. Even today with the latest generation Itanium2 I believe Intel have came to market a little bit too late for their HPC customers (SGI). It was a little unfocused at one point where you didn't actually need the IA64 features, but really NEEDED the performance... I suppose SGI saw that and came up with the Altix ICE.

Now when Altix UV hit the market they're back on focus on performance on HPC, but with added muscles (QPI and other features) to advance on HPC. The downside for SGI is that they're not alone on this bandwagon... IBM have their own system designs coming up with UV-like interconnects.

Itanium was (on paper) much better than the original chips turned out to be, and everyone made the switch (i.e. committed to Itanium) when it was still on paper.

One thing that you didn't bring up (people don't think this way much anymore) is the difference between commercial and technical workloads. SPECers and FLOPs people tend to be technical/HPC workloads, whereas the companies you mentioned tend to be commercial (TPS) workloads. Many of the commercial-oriented machines tend to be rather anemic for their generation in SPECint/SPECfp/FLOPS, but are really good at pushing the data out, which is what a number of customers (i.e. banks) need. Add on hardware accelerators where needed (i.e. crypto), and you wind up with something much faster at what you need than a general purpose HPC box.

I think Itanium was supposed to be targeting the "balanced" market, such that it could be used in either place (Altix as well as Integrity), but the IBM z is definitely a commercial machine, and for many years the i was as well (RS64 was commercial-optimized).